Exploring Uncertainty in dMRI Super-resolution via Probabilistic CNNs

Deep learning has shown success in a wide range of medical image transformation problems, such as super-resolution (SR), denoising and image synthesis. However, the highly ill-posed nature of such problems results in inevitable ambiguity in the learning of networks. In addition, the deterministic nature of the existing methods means that they provide no indication of confidence in its prediction, which hinders reliability assessment and forms a significant barrier to adoption in clinical practice.

This talks will focus on our recent work which proposes a probabilistic CNN method for modelling uncertainty in image enhancement problems. We propose to account for intrinsic uncertainty through a per-patch heteroscedastic noise model and for parameter uncertainty through approximate Bayesian inference in the form of variational dropout. We show that the combined benefits of both lead to the state-of-the-art performance super-resolution of diffusion MR brain images. We further show that the method produces tangible benefits in downstream tractography. In addition, the probabilistic nature of the methods naturally confers a mechanism to quantify uncertainty over the super-resolved output. We demonstrate through experiments on both healthy and pathological brains the potential utility of such an uncertainty measure in the risk assessment of the super-resolved images for subsequent clinical use.